travis does not advance to next build stage when fast_finish: true

[maxheld83]

As per this discussion I understand that fast_finish: true is supported for build stages with matrix expansion and allow_failures: .

However, I observed that a later (deploy) stage did not start, before all earlier (test) jobs where finished, even though the missing one(s) were all allowed failures.

Here's the build.

Here's the .travis.yml (please don't laugh at my clumsy R scripts).

# R for travis: see documentation at https://docs.travis-ci.com/user/languages/r

language: r

# matrix ====
r:
  - oldrel
  - release
  - devel
os:
  - linux
  - osx

# general settings ====

warnings_are_errors: true

cache:
  packages: yes
  directories:
    - $HOME/bin

git:
  # accio submodule is used only for deploy, so skip generally
  submodules: false
  depth: 3

# these need not be in pensieve DESCRIPTION, but are still needed for deploy
# when mentioned here, they get cached
r_packages:
  - covr
  - pkgdown
  - rsconnect

notifications:
  email: false


# linux stuff ====

dist: trusty

sudo: false

addons:
  apt:
    packages:
      - pdf2svg
      - libudunits2-dev


# macOS stuff ====

disable_homebrew: false

brew_packages:
  - pandoc

# script is part of base travis R

jobs:
  fast_finish: true
  allow_failures:
    - r: devel
  include:
    - stage: deploy
      os: linux
      r: oldrel
      script:
        # run test coverage
        - Rscript -e 'covr::codecov()'
        # auth machine for github to get accio submodule
        - echo -e "machine github.com\n  login $GITHUB_PAT" >> ~/.netrc
        - git submodule update --init
        - R CMD INSTALL .
        # make website
        - Rscript -e 'pkgdown::build_site()'
        # repeated install is necessary, because packrat otherwise does not know where pensieve came from
        - Rscript -e "devtools::install_github(repo = 'maxheld83/pensieve', force = TRUE, ref = '${TRAVIS_BRANCH}')"
        # auth shiny
        - Rscript -e "rsconnect::setAccountInfo(name='maxheld83', token='${SHINYAPPS_TOKEN}', secret='${SHINYAPPS_SECRET}')"
      deploy:
        # accio
        - provider: script
          script:
          - Rscript -e "rsconnect::deployApp(appDir = 'inst/accio/', appName = 'accio')"
          skip_cleanup: true
          on:
          branch: master
        - provider: script
          script:
            - Rscript -e "rsconnect::deployApp(appDir = 'inst/accio/', appName = 'accio-${TRAVIS_BRANCH}')"
          skip_cleanup: true
          on:
            all_branches: true
            condition: "$TRAVIS_BRANCH != master"
        # site
        - provider: firebase
          token: $FIREBASE_TOKEN
          project: pensieve-169bf
          skip_cleanup: true
          on:
            branch: master
        - provider: pages
          github_token: $GITHUB_PAT
          local_dir: docs/
          skip_cleanup: true
          on:
            branch: dev

[BanzaiMan]

fast_finish is a build level concern, so this behavior seems correct to me.

However, you are correct to point out that currently we do not move to the next stage when the remaining jobs in the present stage are all allowed to fail.

[gaborbernat]

I would just add that it could be worth evaluating the deploy on flags eagerly too. E.g. if I have a build stage that is triggered on a given branch only, no need to wait around the evaluation of that stage.

[stale]

Thanks for contributing to this issue. As it has been 90 days since the last activity, we are automatically closing the issue in 7 days. This is often because the request was already solved in some way and it just wasn't updated or it's no longer applicable. If that's not the case, please respond before the issue is closed, or open a new one after. We'll gladly take a look again! You can read more here: https://blog.travis-ci.com/2018-03-09-closing-old-issues

[peternewman]

I believe the build still currently doesn't move onto deploy as soon as all the jobs which must pass have successfully finished, hence this issue isn't stale.

[jonhoo]

Not sure if this is related, but in https://travis-ci.org/jonhoo/rust-imap/builds/459508822 I cancelled the last job, which is allowed to fail, and the build was not marked as successful? In theory, the job should have been marked as successful once the previous stage completed since fast_finish is on, but that didn't happen either.

[mkurz]

@jonhoo Same here. If we cancel the last job (which is allowed to fail) the whole build is doomed - however it should be marked successful because every other job was green.
Maybe we should open a new issue for that wrong behaviour?

[jonhoo]

@mkurz filed as #10356

[DrTorte]

Hello,

We're moving our community communications to the Travis CI Community Forum and will no longer be supporting GitHub issues. If this hasn't been solved, please go ahead and create a new thread on the Forum and we'll be better able to help you there. We're looking forward to seeing you there.

If you need support, you can reach our support team here: support@travis-ci.com.

Thanks and happy building!

[jonhoo]

I think https://travis-ci.community/t/work-out-kinks-in-interactions-between-stages-allow-fail-and-fast-finish/1090 is the closest community post.